Semiconductors: The AI Buildout Has No Ceiling Yet

Broadcom crossed into trillion-dollar territory on the back of custom AI chip demand, but the deeper story in semiconductors runs upstream. Every advanced chip — whether it's a GPU, a networking ASIC, or a memory die — requires lithography, etch, deposition, and inspection equipment before a single transistor fires. That supply chain is where the structural leverage lives, and it's where the theme's most durable beneficiaries cluster.

The 83-company graph here spans the full stack: foundries, fabless designers, equipment vendors, materials suppliers, and electronics manufacturers. Not all of them move together. When chip demand softens, fabless names feel it first. Equipment and process-control companies feel it later, because fabs keep spending on yield improvement even when utilization drops.

Equipment Makers Capture the Toll-Road Position

ASML is the only company on earth that makes extreme ultraviolet lithography machines. That monopoly on the most critical step in advanced chip manufacturing gives it pricing power that no chipmaker can replicate. Lam Research and Applied Materials own the etch and deposition steps that follow — also highly concentrated markets. These three companies don't just benefit from AI spending; they are a prerequisite for it. No new fab capacity gets built without them.

KLA Corp sits at the inspection and process control layer. As node sizes shrink below 3nm, defect rates become existential for fabs. KLA's tools catch those defects before they become expensive scrap. That makes its revenue less cyclical than most of the industry — fabs cut equipment budgets before they cut process control.

Materials and Connectivity Are the Underappreciated Lever

Entegris supplies the ultra-pure chemicals and materials that advanced nodes require. Contamination at 2nm can destroy an entire wafer run, so fabs don't shop on price. Amphenol and Coherent represent a different angle: the connectors and optical components that move data between chips and across data center infrastructure. As AI clusters scale, interconnect bandwidth becomes the bottleneck, and both companies are direct beneficiaries of that constraint.

Teradyne tests chips after fabrication — another step that can't be skipped as complexity rises. More transistors per die means more test time per chip, which mechanically drives Teradyne's revenue higher as the industry advances.

The Cycle Risk Is Real, But the Secular Trend Is Louder

Semiconductors are cyclical. Inventory corrections hit hard and fast, as the 2022–2023 downturn demonstrated. United Microelectronics, which serves mature nodes rather than bleeding-edge processes, is more exposed to those swings because its customers have more substitution options and less urgency to lock in capacity.

The secular case, though, is intact. AI training and inference require orders of magnitude more compute than prior workloads. Every major hyperscaler is building or expanding its own silicon program. That creates sustained demand not just for leading-edge logic, but for the HBM memory, advanced packaging, and optical interconnects that surround it. The ETF library captures this theme in diversified form for investors who want exposure without single-stock concentration.

The companies with the most defensible positions are the ones where switching costs are highest and competition is thinnest — process control, lithography, specialty materials. That's where the theme's strongest risk-adjusted returns have historically come from, and the AI buildout gives those companies a longer runway than most cycles provide.